Skip to content

Use case: incident timelines (what was seen vs what was true) #31

Closed

Description

@adityamukho

Summary

Add a use case: deploys, config changes, feature flags and alerts as facts, for incident response and postmortems.

The idea

Store operational events as they arrive. Logs often arrive late, and early reports often get corrected. The two time axes keep both stories:

  • Valid time: when something actually happened.
  • Transaction time: when the team learned about it.

Questions it answers

  • What did the on-call engineer see at 14:02?
  • What was actually true at 14:02, now that late data has arrived?
  • Which config and flag values were live for this service when the error rate rose?

These two answers differ in most real incidents. Most tools keep only one of them.

Why Minigraf fits

  • The seen vs true split is exactly bi-temporal data.
  • Services, deploys, configs and alerts link into a graph, so "what changed upstream of this service?" is a traversal.
  • One file per incident is easy to attach to a postmortem.

Suggested content

  • The idea and the seen vs true example
  • A fact schema (service, deploy, config value, flag, alert, depends-on)
  • One query at a transaction time, one at a valid time, showing the difference
  • A traversal for changes upstream of an affected service
  • A note that raw logs and metrics belong in their own stores; only events go in

Notes


Generated by Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    documentationImprovements or additions to documentation

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions